Contributed to backend reliability and automation by developing and refining features in Python for langgenius/dify and modelscope/ms-swift. Delivered post-publish file parameter support in WorkflowTool, introducing a mechanism to update file mappings based on transfer methods and ensuring correct file ID references after workflow publication. This enhancement reduced post-publish errors and improved workflow automation. Additionally, addressed a critical bug in the EarlyStopping plugin for modelscope/ms-swift, ensuring the patience counter resets correctly on metric improvement to prevent premature training termination. Work focused on API development, plugin maintenance, and machine learning, emphasizing robust backend solutions and improved resource efficiency in model training.